【作者】 吕新荣；

【导师】 高新波；

【作者基本信息】 西安电子科技大学 ， 模式识别与智能系统， 2009， 博士

【摘要】 心脑血管系统疾病已经成为当今世界上发病率、死亡率最高的疾病之一。计算机断层扫描与核磁共振成像等技术的出现使现代影像学检查技术成为血管疾病诊断的重要手段。在血管分析诊断系统中,血管的分割、提取与可视化技术发挥着极其重要的作用。目前的血管分割和提取算法因成像模式、应用领域以及其他一些因素的影响而各不相同,尚无任何一种通用的分割算法能够适用于所有的医学影像模式。在手术过程中,血管组织的可视化能够精确定位血管内部狭窄位置,为手术计划的定制提供依据。虽然经历了二十多年的发展,但是血管可视化技术远未成熟,在理论和实际应用中还有许多亟待解决的问题。本文在目前存在的血管提取与可视化技术的基础上,以计算机断层扫描血管造影体数据作为研究对象,分别利用图形学理论和图像处理技术构建更精确的可视化技术和中心路径提取算法,并将两者进行结合,以便提供更多的血管内外部信息,为血管疾病的诊断提供足够的依据。本文所取得的主要研究成果和创新点如下:(1)针对目前在曲面重建过程中采用多折线段模拟曲线带来的误差问题,根据任意设置的采样步长,采用B样条的方法将所有控制点拟合成曲线,实现了医学图像体数据的高精度曲面重建;并在此基础上引入一些曲面重建的辅助功能,如窗宽窗位调整、旋转曲面重建等,提升曲面重建的性能;(2)为了能够在血管中更精确地观察病变,提出一种基于Snake模型的血管中心路径提取方法。首先,由用户手动在某一层上确定一个血管目标区域,并用初始化轮廓方法确定一个目标轮廓作为Snake模型的初始轮廓,加快了Snake模型的收敛速度;其次,求得该层Snake模型收敛的最终轮廓的中心点作为该层血管的中心点,并将该层的最终轮廓作为下一层的手动轮廓,以此类推,逐层取得血管中心点并连成中心路径;(3)针对目前利用遍历方法提取血管速度较慢的缺点,引入八叉树分解方法以加速血管分割和血管边界距离场的计算,然后建立基于边界距离场的血管组织最大生成树,并对感兴趣的血管分支提取树的主干,即该分支的中心路径,最后用基于图形硬件处理器的三维纹理体绘制方法沿着中心路径实现血管虚拟内窥镜;(4)针对传统边界距离场的计算以分割后的二值体数据为目标而丧失了原有数据特征的缺点,提出一种改进的基于距离场的中心路径提取算法。该算法利用原血管造影体数据中血管体素的梯度值倒数和拉普拉斯变换值作为边界距离场计算的初始值,并用重心法修正基于距离场提取的中心路径。与传统基于边界距离场的提取算法相比,该方法提取的血管中心路径更接近于实际应用中人工提取的路径;(5)为了更准确的提取三维医学体数据中管状器官的中心路径,提出一种基于Hessian矩阵的中心路径提取算法。首先,计算输入的分割后二值数据中每个血管体素的Hessian矩阵;其次,利用每个血管体素Hessian矩阵的特征值和特征向量提取一条粗糙的中心路径;最后,对该中心路径上的每一点,利用尺度空间分析法在其所在的管腔横截面内进行修正,最终得到一条精确的中心路径。上述研究成果分别从血管的可视化与血管的中心路径提取等方面给出了具体的研究方案和实验结果,为血管疾病的诊断提供了更为先进的辅助手段。此外,文中提出的曲面重建方法和不同的中心路径提取算法具有一定的通用性,为管腔可视化技术的理论研究与应用推广提供了新的思路。更多还原

【Abstract】 Cardiovascular and cerebrovascular diseases have become one of the diseases with highest incidence and mortality rates. Computed tomography and magnetic resonance imaging techniques make the modern imaging technology become an important diagnosis means of blood vessel diseases. In the blood vessel analysis and diagnosis system, the extraction and visualization of blood vessels play a vita role. The segmentation and extraction algorithms vary depending on imaging modality, application field and other factors. However, there is no such a general segmentation algorithm that is suitable to all medical imaging modalities. In the blood vessel surgery, visualization can locate the stenosis position which is helpful for doctors making an appropriate surgery plan.After twenty-year development, blood vessel visualization is still in its infancy and remains a great number of problems demanding prompt solution in both theory and application. Based on the existent vessel extraction and visualization techniques, this paper focuses on the research of computed tomography angiography vessel volume datasets. Computer graphics theory and image processing theory are utilized to realize more accurate visualization techniques and centerline extraction algorithms, respectively. The visualization and centerline extracted are also combined to provide more information for the diagnosis of blood vessel diseases. The main contributions and innovative ideas of this paper are summarized as follows:(1) A high-precision curved planar reformation is realized against the errors caused by simulating curves using pieces of lines in curved planar reformation at present. In the high-precision method, all control points are used to produce curves with B-spline fitting according to the sampling steps specified. Based on this, some assistant functions of curved planar reformation are introduced to enhance the performance of the curved planar reformation, such as adjustments of window widths and window levels, rotating curved planar reformation.(2) In order to make more accurate observations of the diseases in blood vessel, a centerline extraction method based on Snake model is proposed. First, a region of interest of blood vessel is determined by users in any slice of the volume data. And an initializing contour method is used to produce the initial contour for Snake model, which accelerates the convergence of the Snake model. Then, a final contour can be generated by the convergence of Snake model and a center point of the contour is obtained as the center of blood vessel in current slice. And the final contour in current slice is regarded as the initial contour of the next slice. In this way, a centerline can be obtained by connecting all the center points slice by slice.(3) Aiming at the low speed in segmentation of blood vessel volume data using traversal method, an octree structure is introduced to accelerate the blood vessel segmentation process and the computation of distance from boundary field. Then, a maximum spanning tree of vascular structure is constructed based on the distance from boundary field and the trunk of the tree is extracted which is the centerline. Finally, a graphics processing unit-based 3D texture-mapping volume rendering method is utilized to show the virtual endoscopy of blood vessel along the centerline.(4) To overcome the disadvantage in computation of the traditional distance from boundary field which loses the features of original data, an improved centerline extraction algorithm based on the distance from boundary field is proposed. In this algorithm, for each target voxel in original computed tomography angiography volume datasets, the sum of reverse of gradient and laplacian transformation value is regarded as the initial value, and the centerline extracted based on distance from boundary field is modified by center of gravity method. Compared to the traditional centerline extraction based on distance from boundary field, the centerline extracted by this algorithm is closer to the one extracted manually.(5) In order to extract more accurate centerlines for vascular structures in 3D medical volume data, a centerline extraction method based on Hessian matrix is proposed. Firstly, Hessian matrix is computed for each target voxel in the segmented data which is always binary. Then, a coarse centerline is generated according to the eigenvalues and the eigenvectors of each target voxel’s Hessian matrix. Finally, scale space analysis method is used to refine each point in the coarse centerline in its cross-section plane. Thus, an accurate centerline is obtained.All achievements above have presented research schemes and experimental results in blood vessel visualization and extraction, which will provide more advanced assistant methods for blood vessel diseases diagnosis. Besides, curved planar reformation and several different centerline extraction algorithms proposed in this paper have generality and enrich the theories and applications of vascular structures visualization.更多还原